SpecOPD: Branch-Parallel On-Policy Distillation for Speculative Decoding
SpecOPD trains speculative decoding drafters on their own sampled histories while parallelizing target supervision, improving acceptance length and training throughput.
ls ./papers | grep "#On-policy distillation"
Trending AI research papers from Hugging Face, summarized in plain language: what each paper does, why it matters and where to read more.
3 papers · #On-policy distillation
SpecOPD trains speculative decoding drafters on their own sampled histories while parallelizing target supervision, improving acceptance length and training throughput.
DecisionRank distills a generative LLM’s sequential ranking behavior into a single-pass reranker, reporting strong retrieval quality and major latency gains.
Youtu-Parsing-Omni combines a shared multimodal encoder with one grounded schema, delivering strong document parsing and notable gains on audio and video.